2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
11 #include "string-list.h"
17 #include "run-command.h"
20 #include "tree-walk.h"
22 #include "pack-revindex.h"
23 #include "sha1-lookup.h"
24 #include "bulk-checkin.h"
25 #include "repository.h"
26 #include "streaming.h"
29 #include "mergesort.h"
32 #include "fetch-object.h"
33 #include "object-store.h"
35 const unsigned char null_sha1[GIT_MAX_RAWSZ];
36 const struct object_id null_oid;
37 const struct object_id empty_tree_oid = {
38 EMPTY_TREE_SHA1_BIN_LITERAL
40 const struct object_id empty_blob_oid = {
41 EMPTY_BLOB_SHA1_BIN_LITERAL
44 static void git_hash_sha1_init(void *ctx)
46 git_SHA1_Init((git_SHA_CTX *)ctx);
49 static void git_hash_sha1_update(void *ctx, const void *data, size_t len)
51 git_SHA1_Update((git_SHA_CTX *)ctx, data, len);
54 static void git_hash_sha1_final(unsigned char *hash, void *ctx)
56 git_SHA1_Final(hash, (git_SHA_CTX *)ctx);
59 static void git_hash_unknown_init(void *ctx)
61 die("trying to init unknown hash");
64 static void git_hash_unknown_update(void *ctx, const void *data, size_t len)
66 die("trying to update unknown hash");
69 static void git_hash_unknown_final(unsigned char *hash, void *ctx)
71 die("trying to finalize unknown hash");
74 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
81 git_hash_unknown_init,
82 git_hash_unknown_update,
83 git_hash_unknown_final,
89 /* "sha1", big-endian */
103 * This is meant to hold a *small* number of objects that you would
104 * want read_sha1_file() to be able to return, but yet you do not want
105 * to write them into the object store (e.g. a browse-only
108 static struct cached_object {
109 unsigned char sha1[20];
110 enum object_type type;
114 static int cached_object_nr, cached_object_alloc;
116 static struct cached_object empty_tree = {
117 EMPTY_TREE_SHA1_BIN_LITERAL,
123 static struct cached_object *find_cached_object(const unsigned char *sha1)
126 struct cached_object *co = cached_objects;
128 for (i = 0; i < cached_object_nr; i++, co++) {
129 if (!hashcmp(co->sha1, sha1))
132 if (!hashcmp(sha1, empty_tree.sha1))
138 static int get_conv_flags(unsigned flags)
140 if (flags & HASH_RENORMALIZE)
141 return CONV_EOL_RENORMALIZE;
142 else if (flags & HASH_WRITE_OBJECT)
143 return global_conv_flags_eol;
149 int mkdir_in_gitdir(const char *path)
151 if (mkdir(path, 0777)) {
152 int saved_errno = errno;
154 struct strbuf sb = STRBUF_INIT;
159 * Are we looking at a path in a symlinked worktree
160 * whose original repository does not yet have it?
161 * e.g. .git/rr-cache pointing at its original
162 * repository in which the user hasn't performed any
163 * conflict resolution yet?
165 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
166 strbuf_readlink(&sb, path, st.st_size) ||
167 !is_absolute_path(sb.buf) ||
168 mkdir(sb.buf, 0777)) {
175 return adjust_shared_perm(path);
178 enum scld_error safe_create_leading_directories(char *path)
180 char *next_component = path + offset_1st_component(path);
181 enum scld_error ret = SCLD_OK;
183 while (ret == SCLD_OK && next_component) {
185 char *slash = next_component, slash_character;
187 while (*slash && !is_dir_sep(*slash))
193 next_component = slash + 1;
194 while (is_dir_sep(*next_component))
196 if (!*next_component)
199 slash_character = *slash;
201 if (!stat(path, &st)) {
203 if (!S_ISDIR(st.st_mode)) {
207 } else if (mkdir(path, 0777)) {
208 if (errno == EEXIST &&
209 !stat(path, &st) && S_ISDIR(st.st_mode))
210 ; /* somebody created it since we checked */
211 else if (errno == ENOENT)
213 * Either mkdir() failed because
214 * somebody just pruned the containing
215 * directory, or stat() failed because
216 * the file that was in our way was
217 * just removed. Either way, inform
218 * the caller that it might be worth
224 } else if (adjust_shared_perm(path)) {
227 *slash = slash_character;
232 enum scld_error safe_create_leading_directories_const(const char *path)
235 /* path points to cache entries, so xstrdup before messing with it */
236 char *buf = xstrdup(path);
237 enum scld_error result = safe_create_leading_directories(buf);
245 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
248 * The number of times we will try to remove empty directories
249 * in the way of path. This is only 1 because if another
250 * process is racily creating directories that conflict with
251 * us, we don't want to fight against them.
253 int remove_directories_remaining = 1;
256 * The number of times that we will try to create the
257 * directories containing path. We are willing to attempt this
258 * more than once, because another process could be trying to
259 * clean up empty directories at the same time as we are
260 * trying to create them.
262 int create_directories_remaining = 3;
264 /* A scratch copy of path, filled lazily if we need it: */
265 struct strbuf path_copy = STRBUF_INIT;
278 if (errno == EISDIR && remove_directories_remaining-- > 0) {
280 * A directory is in the way. Maybe it is empty; try
284 strbuf_addstr(&path_copy, path);
286 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
288 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
290 * Maybe the containing directory didn't exist, or
291 * maybe it was just deleted by a process that is
292 * racing with us to clean up empty directories. Try
295 enum scld_error scld_result;
298 strbuf_addstr(&path_copy, path);
301 scld_result = safe_create_leading_directories(path_copy.buf);
302 if (scld_result == SCLD_OK)
304 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
308 strbuf_release(&path_copy);
313 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
316 for (i = 0; i < 20; i++) {
317 static char hex[] = "0123456789abcdef";
318 unsigned int val = sha1[i];
319 strbuf_addch(buf, hex[val >> 4]);
320 strbuf_addch(buf, hex[val & 0xf]);
322 strbuf_addch(buf, '/');
326 void sha1_file_name(struct repository *r, struct strbuf *buf, const unsigned char *sha1)
328 strbuf_addstr(buf, r->objects->objectdir);
329 strbuf_addch(buf, '/');
330 fill_sha1_path(buf, sha1);
333 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
335 strbuf_setlen(&alt->scratch, alt->base_len);
336 return &alt->scratch;
339 static const char *alt_sha1_path(struct alternate_object_database *alt,
340 const unsigned char *sha1)
342 struct strbuf *buf = alt_scratch_buf(alt);
343 fill_sha1_path(buf, sha1);
348 * Return non-zero iff the path is usable as an alternate object database.
350 static int alt_odb_usable(struct raw_object_store *o,
352 const char *normalized_objdir)
354 struct alternate_object_database *alt;
356 /* Detect cases where alternate disappeared */
357 if (!is_directory(path->buf)) {
358 error("object directory %s does not exist; "
359 "check .git/objects/info/alternates.",
365 * Prevent the common mistake of listing the same
366 * thing twice, or object directory itself.
368 for (alt = o->alt_odb_list; alt; alt = alt->next) {
369 if (!fspathcmp(path->buf, alt->path))
372 if (!fspathcmp(path->buf, normalized_objdir))
379 * Prepare alternate object database registry.
381 * The variable alt_odb_list points at the list of struct
382 * alternate_object_database. The elements on this list come from
383 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
384 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
385 * whose contents is similar to that environment variable but can be
386 * LF separated. Its base points at a statically allocated buffer that
387 * contains "/the/directory/corresponding/to/.git/objects/...", while
388 * its name points just after the slash at the end of ".git/objects/"
389 * in the example above, and has enough space to hold 40-byte hex
390 * SHA1, an extra slash for the first level indirection, and the
393 static void read_info_alternates(struct repository *r,
394 const char *relative_base,
396 static int link_alt_odb_entry(struct repository *r, const char *entry,
397 const char *relative_base, int depth, const char *normalized_objdir)
399 struct alternate_object_database *ent;
400 struct strbuf pathbuf = STRBUF_INIT;
402 if (!is_absolute_path(entry) && relative_base) {
403 strbuf_realpath(&pathbuf, relative_base, 1);
404 strbuf_addch(&pathbuf, '/');
406 strbuf_addstr(&pathbuf, entry);
408 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
409 error("unable to normalize alternate object path: %s",
411 strbuf_release(&pathbuf);
416 * The trailing slash after the directory name is given by
417 * this function at the end. Remove duplicates.
419 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
420 strbuf_setlen(&pathbuf, pathbuf.len - 1);
422 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
423 strbuf_release(&pathbuf);
427 ent = alloc_alt_odb(pathbuf.buf);
429 /* add the alternate entry */
430 *r->objects->alt_odb_tail = ent;
431 r->objects->alt_odb_tail = &(ent->next);
434 /* recursively add alternates */
435 read_info_alternates(r, pathbuf.buf, depth + 1);
437 strbuf_release(&pathbuf);
441 static const char *parse_alt_odb_entry(const char *string,
449 if (*string == '#') {
450 /* comment; consume up to next separator */
451 end = strchrnul(string, sep);
452 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
454 * quoted path; unquote_c_style has copied the
455 * data for us and set "end". Broken quoting (e.g.,
456 * an entry that doesn't end with a quote) falls
457 * back to the unquoted case below.
460 /* normal, unquoted path */
461 end = strchrnul(string, sep);
462 strbuf_add(out, string, end - string);
470 static void link_alt_odb_entries(struct repository *r, const char *alt,
471 int sep, const char *relative_base, int depth)
473 struct strbuf objdirbuf = STRBUF_INIT;
474 struct strbuf entry = STRBUF_INIT;
480 error("%s: ignoring alternate object stores, nesting too deep.",
485 strbuf_add_absolute_path(&objdirbuf, r->objects->objectdir);
486 if (strbuf_normalize_path(&objdirbuf) < 0)
487 die("unable to normalize object directory: %s",
491 alt = parse_alt_odb_entry(alt, sep, &entry);
494 link_alt_odb_entry(r, entry.buf,
495 relative_base, depth, objdirbuf.buf);
497 strbuf_release(&entry);
498 strbuf_release(&objdirbuf);
501 static void read_info_alternates(struct repository *r,
502 const char *relative_base,
506 struct strbuf buf = STRBUF_INIT;
508 path = xstrfmt("%s/info/alternates", relative_base);
509 if (strbuf_read_file(&buf, path, 1024) < 0) {
510 warn_on_fopen_errors(path);
515 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
516 strbuf_release(&buf);
520 struct alternate_object_database *alloc_alt_odb(const char *dir)
522 struct alternate_object_database *ent;
524 FLEX_ALLOC_STR(ent, path, dir);
525 strbuf_init(&ent->scratch, 0);
526 strbuf_addf(&ent->scratch, "%s/", dir);
527 ent->base_len = ent->scratch.len;
532 void add_to_alternates_file(const char *reference)
534 struct lock_file lock = LOCK_INIT;
535 char *alts = git_pathdup("objects/info/alternates");
539 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
540 out = fdopen_lock_file(&lock, "w");
542 die_errno("unable to fdopen alternates lockfile");
544 in = fopen(alts, "r");
546 struct strbuf line = STRBUF_INIT;
548 while (strbuf_getline(&line, in) != EOF) {
549 if (!strcmp(reference, line.buf)) {
553 fprintf_or_die(out, "%s\n", line.buf);
556 strbuf_release(&line);
559 else if (errno != ENOENT)
560 die_errno("unable to read alternates file");
563 rollback_lock_file(&lock);
565 fprintf_or_die(out, "%s\n", reference);
566 if (commit_lock_file(&lock))
567 die_errno("unable to move new alternates file into place");
568 if (the_repository->objects->alt_odb_tail)
569 link_alt_odb_entries(the_repository, reference,
575 void add_to_alternates_memory(const char *reference)
578 * Make sure alternates are initialized, or else our entry may be
579 * overwritten when they are.
581 prepare_alt_odb(the_repository);
583 link_alt_odb_entries(the_repository, reference,
588 * Compute the exact path an alternate is at and returns it. In case of
589 * error NULL is returned and the human readable error is added to `err`
590 * `path` may be relative and should point to $GITDIR.
591 * `err` must not be null.
593 char *compute_alternate_path(const char *path, struct strbuf *err)
595 char *ref_git = NULL;
596 const char *repo, *ref_git_s;
599 ref_git_s = real_path_if_valid(path);
602 strbuf_addf(err, _("path '%s' does not exist"), path);
606 * Beware: read_gitfile(), real_path() and mkpath()
607 * return static buffer
609 ref_git = xstrdup(ref_git_s);
611 repo = read_gitfile(ref_git);
613 repo = read_gitfile(mkpath("%s/.git", ref_git));
616 ref_git = xstrdup(repo);
619 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
620 char *ref_git_git = mkpathdup("%s/.git", ref_git);
622 ref_git = ref_git_git;
623 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
624 struct strbuf sb = STRBUF_INIT;
626 if (get_common_dir(&sb, ref_git)) {
628 _("reference repository '%s' as a linked "
629 "checkout is not supported yet."),
634 strbuf_addf(err, _("reference repository '%s' is not a "
635 "local repository."), path);
639 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
640 strbuf_addf(err, _("reference repository '%s' is shallow"),
646 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
648 _("reference repository '%s' is grafted"),
656 FREE_AND_NULL(ref_git);
662 int foreach_alt_odb(alt_odb_fn fn, void *cb)
664 struct alternate_object_database *ent;
667 prepare_alt_odb(the_repository);
668 for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
676 void prepare_alt_odb(struct repository *r)
678 if (r->objects->alt_odb_tail)
681 r->objects->alt_odb_tail = &r->objects->alt_odb_list;
682 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
684 read_info_alternates(r, r->objects->objectdir, 0);
687 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
688 static int freshen_file(const char *fn)
691 t.actime = t.modtime = time(NULL);
692 return !utime(fn, &t);
696 * All of the check_and_freshen functions return 1 if the file exists and was
697 * freshened (if freshening was requested), 0 otherwise. If they return
698 * 0, you should not assume that it is safe to skip a write of the object (it
699 * either does not exist on disk, or has a stale mtime and may be subject to
702 int check_and_freshen_file(const char *fn, int freshen)
704 if (access(fn, F_OK))
706 if (freshen && !freshen_file(fn))
711 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
713 static struct strbuf buf = STRBUF_INIT;
716 sha1_file_name(the_repository, &buf, sha1);
718 return check_and_freshen_file(buf.buf, freshen);
721 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
723 struct alternate_object_database *alt;
724 prepare_alt_odb(the_repository);
725 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
726 const char *path = alt_sha1_path(alt, sha1);
727 if (check_and_freshen_file(path, freshen))
733 static int check_and_freshen(const unsigned char *sha1, int freshen)
735 return check_and_freshen_local(sha1, freshen) ||
736 check_and_freshen_nonlocal(sha1, freshen);
739 int has_loose_object_nonlocal(const unsigned char *sha1)
741 return check_and_freshen_nonlocal(sha1, 0);
744 static int has_loose_object(const unsigned char *sha1)
746 return check_and_freshen(sha1, 0);
749 static void mmap_limit_check(size_t length)
751 static size_t limit = 0;
753 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
758 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
759 (uintmax_t)length, (uintmax_t)limit);
762 void *xmmap_gently(void *start, size_t length,
763 int prot, int flags, int fd, off_t offset)
767 mmap_limit_check(length);
768 ret = mmap(start, length, prot, flags, fd, offset);
769 if (ret == MAP_FAILED) {
772 release_pack_memory(length);
773 ret = mmap(start, length, prot, flags, fd, offset);
778 void *xmmap(void *start, size_t length,
779 int prot, int flags, int fd, off_t offset)
781 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
782 if (ret == MAP_FAILED)
783 die_errno("mmap failed");
788 * With an in-core object data in "map", rehash it to make sure the
789 * object name actually matches "sha1" to detect object corruption.
790 * With "map" == NULL, try reading the object named with "sha1" using
791 * the streaming interface and rehash it to do the same.
793 int check_sha1_signature(const unsigned char *sha1, void *map,
794 unsigned long size, const char *type)
796 unsigned char real_sha1[20];
797 enum object_type obj_type;
798 struct git_istream *st;
804 hash_sha1_file(map, size, type, real_sha1);
805 return hashcmp(sha1, real_sha1) ? -1 : 0;
808 st = open_istream(sha1, &obj_type, &size, NULL);
812 /* Generate the header */
813 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
817 git_SHA1_Update(&c, hdr, hdrlen);
820 ssize_t readlen = read_istream(st, buf, sizeof(buf));
828 git_SHA1_Update(&c, buf, readlen);
830 git_SHA1_Final(real_sha1, &c);
832 return hashcmp(sha1, real_sha1) ? -1 : 0;
835 int git_open_cloexec(const char *name, int flags)
838 static int o_cloexec = O_CLOEXEC;
840 fd = open(name, flags | o_cloexec);
841 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
842 /* Try again w/o O_CLOEXEC: the kernel might not support it */
843 o_cloexec &= ~O_CLOEXEC;
844 fd = open(name, flags | o_cloexec);
847 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
849 static int fd_cloexec = FD_CLOEXEC;
851 if (!o_cloexec && 0 <= fd && fd_cloexec) {
852 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
853 int flags = fcntl(fd, F_GETFD);
854 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
863 * Find "sha1" as a loose object in the local repository or in an alternate.
864 * Returns 0 on success, negative on failure.
866 * The "path" out-parameter will give the path of the object we found (if any).
867 * Note that it may point to static storage and is only valid until another
868 * call to sha1_file_name(), etc.
870 static int stat_sha1_file(struct repository *r, const unsigned char *sha1,
871 struct stat *st, const char **path)
873 struct alternate_object_database *alt;
874 static struct strbuf buf = STRBUF_INIT;
877 sha1_file_name(r, &buf, sha1);
880 if (!lstat(*path, st))
885 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
886 *path = alt_sha1_path(alt, sha1);
887 if (!lstat(*path, st))
895 * Like stat_sha1_file(), but actually open the object and return the
896 * descriptor. See the caveats on the "path" parameter above.
898 static int open_sha1_file(struct repository *r,
899 const unsigned char *sha1, const char **path)
902 struct alternate_object_database *alt;
903 int most_interesting_errno;
904 static struct strbuf buf = STRBUF_INIT;
907 sha1_file_name(r, &buf, sha1);
910 fd = git_open(*path);
913 most_interesting_errno = errno;
916 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
917 *path = alt_sha1_path(alt, sha1);
918 fd = git_open(*path);
921 if (most_interesting_errno == ENOENT)
922 most_interesting_errno = errno;
924 errno = most_interesting_errno;
929 * Map the loose object at "path" if it is not NULL, or the path found by
930 * searching for a loose object named "sha1".
932 #define map_sha1_file_1(r, p, s, si) map_sha1_file_1_##r(p, s, si)
933 static void *map_sha1_file_1_the_repository(const char *path,
934 const unsigned char *sha1,
943 fd = open_sha1_file(the_repository, sha1, &path);
948 if (!fstat(fd, &st)) {
949 *size = xsize_t(st.st_size);
951 /* mmap() is forbidden on empty files */
952 error("object file %s is empty", path);
955 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
962 void *map_sha1_file_the_repository(const unsigned char *sha1, unsigned long *size)
964 return map_sha1_file_1(the_repository, NULL, sha1, size);
967 static int unpack_sha1_short_header(git_zstream *stream,
968 unsigned char *map, unsigned long mapsize,
969 void *buffer, unsigned long bufsiz)
971 /* Get the data stream */
972 memset(stream, 0, sizeof(*stream));
973 stream->next_in = map;
974 stream->avail_in = mapsize;
975 stream->next_out = buffer;
976 stream->avail_out = bufsiz;
978 git_inflate_init(stream);
979 return git_inflate(stream, 0);
982 int unpack_sha1_header(git_zstream *stream,
983 unsigned char *map, unsigned long mapsize,
984 void *buffer, unsigned long bufsiz)
986 int status = unpack_sha1_short_header(stream, map, mapsize,
992 /* Make sure we have the terminating NUL */
993 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
998 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
999 unsigned long mapsize, void *buffer,
1000 unsigned long bufsiz, struct strbuf *header)
1004 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1009 * Check if entire header is unpacked in the first iteration.
1011 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1015 * buffer[0..bufsiz] was not large enough. Copy the partial
1016 * result out to header, and then append the result of further
1017 * reading the stream.
1019 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1020 stream->next_out = buffer;
1021 stream->avail_out = bufsiz;
1024 status = git_inflate(stream, 0);
1025 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1026 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1028 stream->next_out = buffer;
1029 stream->avail_out = bufsiz;
1030 } while (status != Z_STREAM_END);
1034 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1036 int bytes = strlen(buffer) + 1;
1037 unsigned char *buf = xmallocz(size);
1041 n = stream->total_out - bytes;
1044 memcpy(buf, (char *) buffer + bytes, n);
1046 if (bytes <= size) {
1048 * The above condition must be (bytes <= size), not
1049 * (bytes < size). In other words, even though we
1050 * expect no more output and set avail_out to zero,
1051 * the input zlib stream may have bytes that express
1052 * "this concludes the stream", and we *do* want to
1055 * Otherwise we would not be able to test that we
1056 * consumed all the input to reach the expected size;
1057 * we also want to check that zlib tells us that all
1058 * went well with status == Z_STREAM_END at the end.
1060 stream->next_out = buf + bytes;
1061 stream->avail_out = size - bytes;
1062 while (status == Z_OK)
1063 status = git_inflate(stream, Z_FINISH);
1065 if (status == Z_STREAM_END && !stream->avail_in) {
1066 git_inflate_end(stream);
1071 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1072 else if (stream->avail_in)
1073 error("garbage at end of loose object '%s'",
1080 * We used to just use "sscanf()", but that's actually way
1081 * too permissive for what we want to check. So do an anal
1082 * object header parse by hand.
1084 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1087 const char *type_buf = hdr;
1089 int type, type_len = 0;
1092 * The type can be of any size but is followed by
1104 type = type_from_string_gently(type_buf, type_len, 1);
1106 strbuf_add(oi->typename, type_buf, type_len);
1108 * Set type to 0 if its an unknown object and
1109 * we're obtaining the type using '--allow-unknown-type'
1112 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1115 die("invalid object type");
1120 * The length must follow immediately, and be in canonical
1121 * decimal format (ie "010" is not valid).
1123 size = *hdr++ - '0';
1128 unsigned long c = *hdr - '0';
1132 size = size * 10 + c;
1140 * The length must be followed by a zero byte
1142 return *hdr ? -1 : type;
1145 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1147 struct object_info oi = OBJECT_INFO_INIT;
1150 return parse_sha1_header_extended(hdr, &oi, 0);
1153 #define sha1_loose_object_info(r, s, o, f) sha1_loose_object_info_##r(s, o, f)
1154 static int sha1_loose_object_info_the_repository(const unsigned char *sha1,
1155 struct object_info *oi,
1159 unsigned long mapsize;
1163 struct strbuf hdrbuf = STRBUF_INIT;
1164 unsigned long size_scratch;
1166 if (oi->delta_base_sha1)
1167 hashclr(oi->delta_base_sha1);
1170 * If we don't care about type or size, then we don't
1171 * need to look inside the object at all. Note that we
1172 * do not optimize out the stat call, even if the
1173 * caller doesn't care about the disk-size, since our
1174 * return value implicitly indicates whether the
1175 * object even exists.
1177 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1180 if (stat_sha1_file(the_repository, sha1, &st, &path) < 0)
1183 *oi->disk_sizep = st.st_size;
1187 map = map_sha1_file(the_repository, sha1, &mapsize);
1192 oi->sizep = &size_scratch;
1195 *oi->disk_sizep = mapsize;
1196 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1197 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1198 status = error("unable to unpack %s header with --allow-unknown-type",
1200 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1201 status = error("unable to unpack %s header",
1205 else if (hdrbuf.len) {
1206 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1207 status = error("unable to parse %s header with --allow-unknown-type",
1209 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1210 status = error("unable to parse %s header", sha1_to_hex(sha1));
1212 if (status >= 0 && oi->contentp) {
1213 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1215 if (!*oi->contentp) {
1216 git_inflate_end(&stream);
1220 git_inflate_end(&stream);
1222 munmap(map, mapsize);
1223 if (status && oi->typep)
1224 *oi->typep = status;
1225 if (oi->sizep == &size_scratch)
1227 strbuf_release(&hdrbuf);
1228 oi->whence = OI_LOOSE;
1229 return (status < 0) ? status : 0;
1232 int fetch_if_missing = 1;
1234 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1236 static struct object_info blank_oi = OBJECT_INFO_INIT;
1237 struct pack_entry e;
1239 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1240 lookup_replace_object(sha1) :
1242 int already_retried = 0;
1244 if (is_null_sha1(real))
1250 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1251 struct cached_object *co = find_cached_object(real);
1254 *(oi->typep) = co->type;
1256 *(oi->sizep) = co->size;
1258 *(oi->disk_sizep) = 0;
1259 if (oi->delta_base_sha1)
1260 hashclr(oi->delta_base_sha1);
1262 strbuf_addstr(oi->typename, typename(co->type));
1264 *oi->contentp = xmemdupz(co->buf, co->size);
1265 oi->whence = OI_CACHED;
1271 if (find_pack_entry(real, &e))
1274 /* Most likely it's a loose object. */
1275 if (!sha1_loose_object_info(the_repository, real, oi, flags))
1278 /* Not a loose object; someone else may have just packed it. */
1279 reprepare_packed_git();
1280 if (find_pack_entry(real, &e))
1283 /* Check if it is a missing object */
1284 if (fetch_if_missing && repository_format_partial_clone &&
1287 * TODO Investigate haveing fetch_object() return
1288 * TODO error/success and stopping the music here.
1290 fetch_object(repository_format_partial_clone, real);
1291 already_retried = 1;
1298 if (oi == &blank_oi)
1300 * We know that the caller doesn't actually need the
1301 * information below, so return early.
1304 rtype = packed_object_info(e.p, e.offset, oi);
1306 mark_bad_packed_object(e.p, real);
1307 return sha1_object_info_extended(real, oi, 0);
1308 } else if (oi->whence == OI_PACKED) {
1309 oi->u.packed.offset = e.offset;
1310 oi->u.packed.pack = e.p;
1311 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1312 rtype == OBJ_OFS_DELTA);
1318 /* returns enum object_type or negative */
1319 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1321 enum object_type type;
1322 struct object_info oi = OBJECT_INFO_INIT;
1326 if (sha1_object_info_extended(sha1, &oi,
1327 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1332 static void *read_object(const unsigned char *sha1, enum object_type *type,
1333 unsigned long *size)
1335 struct object_info oi = OBJECT_INFO_INIT;
1339 oi.contentp = &content;
1341 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1346 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1347 unsigned char *sha1)
1349 struct cached_object *co;
1351 hash_sha1_file(buf, len, typename(type), sha1);
1352 if (has_sha1_file(sha1) || find_cached_object(sha1))
1354 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1355 co = &cached_objects[cached_object_nr++];
1358 co->buf = xmalloc(len);
1359 memcpy(co->buf, buf, len);
1360 hashcpy(co->sha1, sha1);
1365 * This function dies on corrupt objects; the callers who want to
1366 * deal with them should arrange to call read_object() and give error
1367 * messages themselves.
1369 void *read_sha1_file_extended(const unsigned char *sha1,
1370 enum object_type *type,
1371 unsigned long *size,
1375 const struct packed_git *p;
1378 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1382 data = read_object(repl, type, size);
1386 if (errno && errno != ENOENT)
1387 die_errno("failed to read object %s", sha1_to_hex(sha1));
1389 /* die if we replaced an object with one that does not exist */
1391 die("replacement %s not found for %s",
1392 sha1_to_hex(repl), sha1_to_hex(sha1));
1394 if (!stat_sha1_file(the_repository, repl, &st, &path))
1395 die("loose object %s (stored in %s) is corrupt",
1396 sha1_to_hex(repl), path);
1398 if ((p = has_packed_and_bad(repl)) != NULL)
1399 die("packed object %s (stored in %s) is corrupt",
1400 sha1_to_hex(repl), p->pack_name);
1405 void *read_object_with_reference(const unsigned char *sha1,
1406 const char *required_type_name,
1407 unsigned long *size,
1408 unsigned char *actual_sha1_return)
1410 enum object_type type, required_type;
1412 unsigned long isize;
1413 unsigned char actual_sha1[20];
1415 required_type = type_from_string(required_type_name);
1416 hashcpy(actual_sha1, sha1);
1418 int ref_length = -1;
1419 const char *ref_type = NULL;
1421 buffer = read_sha1_file(actual_sha1, &type, &isize);
1424 if (type == required_type) {
1426 if (actual_sha1_return)
1427 hashcpy(actual_sha1_return, actual_sha1);
1430 /* Handle references */
1431 else if (type == OBJ_COMMIT)
1433 else if (type == OBJ_TAG)
1434 ref_type = "object ";
1439 ref_length = strlen(ref_type);
1441 if (ref_length + 40 > isize ||
1442 memcmp(buffer, ref_type, ref_length) ||
1443 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1448 /* Now we have the ID of the referred-to object in
1449 * actual_sha1. Check again. */
1453 static void write_sha1_file_prepare(const void *buf, unsigned long len,
1454 const char *type, unsigned char *sha1,
1455 char *hdr, int *hdrlen)
1459 /* Generate the header */
1460 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1464 git_SHA1_Update(&c, hdr, *hdrlen);
1465 git_SHA1_Update(&c, buf, len);
1466 git_SHA1_Final(sha1, &c);
1470 * Move the just written object into its final resting place.
1472 int finalize_object_file(const char *tmpfile, const char *filename)
1476 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1478 else if (link(tmpfile, filename))
1482 * Coda hack - coda doesn't like cross-directory links,
1483 * so we fall back to a rename, which will mean that it
1484 * won't be able to check collisions, but that's not a
1487 * The same holds for FAT formatted media.
1489 * When this succeeds, we just return. We have nothing
1492 if (ret && ret != EEXIST) {
1494 if (!rename(tmpfile, filename))
1498 unlink_or_warn(tmpfile);
1500 if (ret != EEXIST) {
1501 return error_errno("unable to write sha1 filename %s", filename);
1503 /* FIXME!!! Collision check here ? */
1507 if (adjust_shared_perm(filename))
1508 return error("unable to set permission to '%s'", filename);
1512 static int write_buffer(int fd, const void *buf, size_t len)
1514 if (write_in_full(fd, buf, len) < 0)
1515 return error_errno("file write error");
1519 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1520 unsigned char *sha1)
1523 int hdrlen = sizeof(hdr);
1524 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1528 /* Finalize a file on disk, and close it. */
1529 static void close_sha1_file(int fd)
1531 if (fsync_object_files)
1532 fsync_or_die(fd, "sha1 file");
1534 die_errno("error when closing sha1 file");
1537 /* Size of directory component, including the ending '/' */
1538 static inline int directory_size(const char *filename)
1540 const char *s = strrchr(filename, '/');
1543 return s - filename + 1;
1547 * This creates a temporary file in the same directory as the final
1550 * We want to avoid cross-directory filename renames, because those
1551 * can have problems on various filesystems (FAT, NFS, Coda).
1553 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1555 int fd, dirlen = directory_size(filename);
1558 strbuf_add(tmp, filename, dirlen);
1559 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1560 fd = git_mkstemp_mode(tmp->buf, 0444);
1561 if (fd < 0 && dirlen && errno == ENOENT) {
1563 * Make sure the directory exists; note that the contents
1564 * of the buffer are undefined after mkstemp returns an
1565 * error, so we have to rewrite the whole buffer from
1569 strbuf_add(tmp, filename, dirlen - 1);
1570 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1572 if (adjust_shared_perm(tmp->buf))
1576 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1577 fd = git_mkstemp_mode(tmp->buf, 0444);
1582 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1583 const void *buf, unsigned long len, time_t mtime)
1586 unsigned char compressed[4096];
1589 unsigned char parano_sha1[20];
1590 static struct strbuf tmp_file = STRBUF_INIT;
1591 static struct strbuf filename = STRBUF_INIT;
1593 strbuf_reset(&filename);
1594 sha1_file_name(the_repository, &filename, sha1);
1596 fd = create_tmpfile(&tmp_file, filename.buf);
1598 if (errno == EACCES)
1599 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1601 return error_errno("unable to create temporary file");
1605 git_deflate_init(&stream, zlib_compression_level);
1606 stream.next_out = compressed;
1607 stream.avail_out = sizeof(compressed);
1610 /* First header.. */
1611 stream.next_in = (unsigned char *)hdr;
1612 stream.avail_in = hdrlen;
1613 while (git_deflate(&stream, 0) == Z_OK)
1615 git_SHA1_Update(&c, hdr, hdrlen);
1617 /* Then the data itself.. */
1618 stream.next_in = (void *)buf;
1619 stream.avail_in = len;
1621 unsigned char *in0 = stream.next_in;
1622 ret = git_deflate(&stream, Z_FINISH);
1623 git_SHA1_Update(&c, in0, stream.next_in - in0);
1624 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1625 die("unable to write sha1 file");
1626 stream.next_out = compressed;
1627 stream.avail_out = sizeof(compressed);
1628 } while (ret == Z_OK);
1630 if (ret != Z_STREAM_END)
1631 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1632 ret = git_deflate_end_gently(&stream);
1634 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1635 git_SHA1_Final(parano_sha1, &c);
1636 if (hashcmp(sha1, parano_sha1) != 0)
1637 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1639 close_sha1_file(fd);
1644 utb.modtime = mtime;
1645 if (utime(tmp_file.buf, &utb) < 0)
1646 warning_errno("failed utime() on %s", tmp_file.buf);
1649 return finalize_object_file(tmp_file.buf, filename.buf);
1652 static int freshen_loose_object(const unsigned char *sha1)
1654 return check_and_freshen(sha1, 1);
1657 static int freshen_packed_object(const unsigned char *sha1)
1659 struct pack_entry e;
1660 if (!find_pack_entry(sha1, &e))
1664 if (!freshen_file(e.p->pack_name))
1670 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1673 int hdrlen = sizeof(hdr);
1675 /* Normally if we have it in the pack then we do not bother writing
1676 * it out into .git/objects/??/?{38} file.
1678 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1679 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1681 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1684 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1685 struct object_id *oid, unsigned flags)
1688 int hdrlen, status = 0;
1690 /* type string, SP, %lu of the length plus NUL must fit this */
1691 hdrlen = strlen(type) + 32;
1692 header = xmalloc(hdrlen);
1693 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1695 if (!(flags & HASH_WRITE_OBJECT))
1697 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1699 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1706 int force_object_loose(const unsigned char *sha1, time_t mtime)
1710 enum object_type type;
1715 if (has_loose_object(sha1))
1717 buf = read_object(sha1, &type, &len);
1719 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1720 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1721 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1727 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1729 if (!startup_info->have_repository)
1731 return sha1_object_info_extended(sha1, NULL,
1732 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1735 int has_object_file(const struct object_id *oid)
1737 return has_sha1_file(oid->hash);
1740 int has_object_file_with_flags(const struct object_id *oid, int flags)
1742 return has_sha1_file_with_flags(oid->hash, flags);
1745 static void check_tree(const void *buf, size_t size)
1747 struct tree_desc desc;
1748 struct name_entry entry;
1750 init_tree_desc(&desc, buf, size);
1751 while (tree_entry(&desc, &entry))
1753 * tree_entry() will die() on malformed entries */
1757 static void check_commit(const void *buf, size_t size)
1760 memset(&c, 0, sizeof(c));
1761 if (parse_commit_buffer(&c, buf, size))
1762 die("corrupt commit");
1765 static void check_tag(const void *buf, size_t size)
1768 memset(&t, 0, sizeof(t));
1769 if (parse_tag_buffer(&t, buf, size))
1773 static int index_mem(struct object_id *oid, void *buf, size_t size,
1774 enum object_type type,
1775 const char *path, unsigned flags)
1777 int ret, re_allocated = 0;
1778 int write_object = flags & HASH_WRITE_OBJECT;
1784 * Convert blobs to git internal format
1786 if ((type == OBJ_BLOB) && path) {
1787 struct strbuf nbuf = STRBUF_INIT;
1788 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1789 get_conv_flags(flags))) {
1790 buf = strbuf_detach(&nbuf, &size);
1794 if (flags & HASH_FORMAT_CHECK) {
1795 if (type == OBJ_TREE)
1796 check_tree(buf, size);
1797 if (type == OBJ_COMMIT)
1798 check_commit(buf, size);
1799 if (type == OBJ_TAG)
1800 check_tag(buf, size);
1804 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1806 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1812 static int index_stream_convert_blob(struct object_id *oid, int fd,
1813 const char *path, unsigned flags)
1816 const int write_object = flags & HASH_WRITE_OBJECT;
1817 struct strbuf sbuf = STRBUF_INIT;
1820 assert(would_convert_to_git_filter_fd(path));
1822 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1823 get_conv_flags(flags));
1826 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1829 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1831 strbuf_release(&sbuf);
1835 static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1836 const char *path, unsigned flags)
1838 struct strbuf sbuf = STRBUF_INIT;
1841 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1842 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1845 strbuf_release(&sbuf);
1849 #define SMALL_FILE_SIZE (32*1024)
1851 static int index_core(struct object_id *oid, int fd, size_t size,
1852 enum object_type type, const char *path,
1858 ret = index_mem(oid, "", size, type, path, flags);
1859 } else if (size <= SMALL_FILE_SIZE) {
1860 char *buf = xmalloc(size);
1861 ssize_t read_result = read_in_full(fd, buf, size);
1862 if (read_result < 0)
1863 ret = error_errno("read error while indexing %s",
1864 path ? path : "<unknown>");
1865 else if (read_result != size)
1866 ret = error("short read while indexing %s",
1867 path ? path : "<unknown>");
1869 ret = index_mem(oid, buf, size, type, path, flags);
1872 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1873 ret = index_mem(oid, buf, size, type, path, flags);
1880 * This creates one packfile per large blob unless bulk-checkin
1881 * machinery is "plugged".
1883 * This also bypasses the usual "convert-to-git" dance, and that is on
1884 * purpose. We could write a streaming version of the converting
1885 * functions and insert that before feeding the data to fast-import
1886 * (or equivalent in-core API described above). However, that is
1887 * somewhat complicated, as we do not know the size of the filter
1888 * result, which we need to know beforehand when writing a git object.
1889 * Since the primary motivation for trying to stream from the working
1890 * tree file and to avoid mmaping it in core is to deal with large
1891 * binary blobs, they generally do not want to get any conversion, and
1892 * callers should avoid this code path when filters are requested.
1894 static int index_stream(struct object_id *oid, int fd, size_t size,
1895 enum object_type type, const char *path,
1898 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1901 int index_fd(struct object_id *oid, int fd, struct stat *st,
1902 enum object_type type, const char *path, unsigned flags)
1907 * Call xsize_t() only when needed to avoid potentially unnecessary
1908 * die() for large files.
1910 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1911 ret = index_stream_convert_blob(oid, fd, path, flags);
1912 else if (!S_ISREG(st->st_mode))
1913 ret = index_pipe(oid, fd, type, path, flags);
1914 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1915 (path && would_convert_to_git(&the_index, path)))
1916 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1919 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1925 int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1928 struct strbuf sb = STRBUF_INIT;
1931 switch (st->st_mode & S_IFMT) {
1933 fd = open(path, O_RDONLY);
1935 return error_errno("open(\"%s\")", path);
1936 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1937 return error("%s: failed to insert into database",
1941 if (strbuf_readlink(&sb, path, st->st_size))
1942 return error_errno("readlink(\"%s\")", path);
1943 if (!(flags & HASH_WRITE_OBJECT))
1944 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1945 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1946 rc = error("%s: failed to insert into database", path);
1947 strbuf_release(&sb);
1950 return resolve_gitlink_ref(path, "HEAD", oid);
1952 return error("%s: unsupported file type", path);
1957 int read_pack_header(int fd, struct pack_header *header)
1959 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1960 /* "eof before pack header was fully read" */
1961 return PH_ERROR_EOF;
1963 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1964 /* "protocol error (pack signature mismatch detected)" */
1965 return PH_ERROR_PACK_SIGNATURE;
1966 if (!pack_version_ok(header->hdr_version))
1967 /* "protocol error (pack version unsupported)" */
1968 return PH_ERROR_PROTOCOL;
1972 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1974 enum object_type type = sha1_object_info(sha1, NULL);
1976 die("%s is not a valid object", sha1_to_hex(sha1));
1978 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1982 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1983 struct strbuf *path,
1984 each_loose_object_fn obj_cb,
1985 each_loose_cruft_fn cruft_cb,
1986 each_loose_subdir_fn subdir_cb,
1989 size_t origlen, baselen;
1993 struct object_id oid;
1995 if (subdir_nr > 0xff)
1996 BUG("invalid loose object subdirectory: %x", subdir_nr);
1998 origlen = path->len;
1999 strbuf_complete(path, '/');
2000 strbuf_addf(path, "%02x", subdir_nr);
2002 dir = opendir(path->buf);
2004 if (errno != ENOENT)
2005 r = error_errno("unable to open %s", path->buf);
2006 strbuf_setlen(path, origlen);
2010 oid.hash[0] = subdir_nr;
2011 strbuf_addch(path, '/');
2012 baselen = path->len;
2014 while ((de = readdir(dir))) {
2016 if (is_dot_or_dotdot(de->d_name))
2019 namelen = strlen(de->d_name);
2020 strbuf_setlen(path, baselen);
2021 strbuf_add(path, de->d_name, namelen);
2022 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2023 !hex_to_bytes(oid.hash + 1, de->d_name,
2024 GIT_SHA1_RAWSZ - 1)) {
2026 r = obj_cb(&oid, path->buf, data);
2034 r = cruft_cb(de->d_name, path->buf, data);
2041 strbuf_setlen(path, baselen - 1);
2042 if (!r && subdir_cb)
2043 r = subdir_cb(subdir_nr, path->buf, data);
2045 strbuf_setlen(path, origlen);
2050 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2051 each_loose_object_fn obj_cb,
2052 each_loose_cruft_fn cruft_cb,
2053 each_loose_subdir_fn subdir_cb,
2059 for (i = 0; i < 256; i++) {
2060 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2069 int for_each_loose_file_in_objdir(const char *path,
2070 each_loose_object_fn obj_cb,
2071 each_loose_cruft_fn cruft_cb,
2072 each_loose_subdir_fn subdir_cb,
2075 struct strbuf buf = STRBUF_INIT;
2078 strbuf_addstr(&buf, path);
2079 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2081 strbuf_release(&buf);
2086 struct loose_alt_odb_data {
2087 each_loose_object_fn *cb;
2091 static int loose_from_alt_odb(struct alternate_object_database *alt,
2094 struct loose_alt_odb_data *data = vdata;
2095 struct strbuf buf = STRBUF_INIT;
2098 strbuf_addstr(&buf, alt->path);
2099 r = for_each_loose_file_in_objdir_buf(&buf,
2100 data->cb, NULL, NULL,
2102 strbuf_release(&buf);
2106 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2108 struct loose_alt_odb_data alt;
2111 r = for_each_loose_file_in_objdir(get_object_directory(),
2112 cb, NULL, NULL, data);
2116 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2121 return foreach_alt_odb(loose_from_alt_odb, &alt);
2124 static int check_stream_sha1(git_zstream *stream,
2128 const unsigned char *expected_sha1)
2131 unsigned char real_sha1[GIT_MAX_RAWSZ];
2132 unsigned char buf[4096];
2133 unsigned long total_read;
2137 git_SHA1_Update(&c, hdr, stream->total_out);
2140 * We already read some bytes into hdr, but the ones up to the NUL
2141 * do not count against the object's content size.
2143 total_read = stream->total_out - strlen(hdr) - 1;
2146 * This size comparison must be "<=" to read the final zlib packets;
2147 * see the comment in unpack_sha1_rest for details.
2149 while (total_read <= size &&
2150 (status == Z_OK || status == Z_BUF_ERROR)) {
2151 stream->next_out = buf;
2152 stream->avail_out = sizeof(buf);
2153 if (size - total_read < stream->avail_out)
2154 stream->avail_out = size - total_read;
2155 status = git_inflate(stream, Z_FINISH);
2156 git_SHA1_Update(&c, buf, stream->next_out - buf);
2157 total_read += stream->next_out - buf;
2159 git_inflate_end(stream);
2161 if (status != Z_STREAM_END) {
2162 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2165 if (stream->avail_in) {
2166 error("garbage at end of loose object '%s'",
2167 sha1_to_hex(expected_sha1));
2171 git_SHA1_Final(real_sha1, &c);
2172 if (hashcmp(expected_sha1, real_sha1)) {
2173 error("sha1 mismatch for %s (expected %s)", path,
2174 sha1_to_hex(expected_sha1));
2181 int read_loose_object(const char *path,
2182 const unsigned char *expected_sha1,
2183 enum object_type *type,
2184 unsigned long *size,
2189 unsigned long mapsize;
2195 map = map_sha1_file_1(the_repository, path, NULL, &mapsize);
2197 error_errno("unable to mmap %s", path);
2201 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2202 error("unable to unpack header of %s", path);
2206 *type = parse_sha1_header(hdr, size);
2208 error("unable to parse header of %s", path);
2209 git_inflate_end(&stream);
2213 if (*type == OBJ_BLOB) {
2214 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2217 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2219 error("unable to unpack contents of %s", path);
2220 git_inflate_end(&stream);
2223 if (check_sha1_signature(expected_sha1, *contents,
2224 *size, typename(*type))) {
2225 error("sha1 mismatch for %s (expected %s)", path,
2226 sha1_to_hex(expected_sha1));
2232 ret = 0; /* everything checks out */
2236 munmap(map, mapsize);